[Weight processing] Copy over `new_data` attributes in `replace_parameter` (#49601)

fxmarty-amd committed 3 months ago

[Attention][MLA] Per-request scheduling for MLA chunked context (#50613)

Matthew Bonanni committed 3 months ago

Fully generalise input embedding handling in Transformers modelling backend (#51247)

Harry Mellor committed 3 months ago

[CI] Run basic fullgraph correctness on one GPU (#51271)

Michael Goin committed 3 months ago

attn_res kernel latency improvements (#50185)

gnovack committed 3 months ago

Update vllm to point to flash-attention commit that builds FA3 with torch stable API. (Retry) (#49599)

Chris Leonard committed 3 months ago

[VocabParallelEmbedding] fix extra_repr fields concat (#51224)

Ning Xie committed 3 months ago

Remove the XPU branch of topk_softplus_sqrt (#51242)

Liangqiusong committed 3 months ago

[Bugfix][KV Offload] Clean up resources after initialization failure (#51227)

AlexHuang committed 3 months ago

[Bugfix][Model] Add missing fused_qkv_a_proj to Kimi-Linear packed_modules_mapping (#51249)

yjz committed 3 months ago

[Rust Frontend] Add standalone Rust renderer (#50289)

Sage committed 3 months ago

[Bugfix][KV Offload] Fall back when MADV_POPULATE_WRITE is unsupported (#51116)

AlexHuang committed 3 months ago

Interns2mobius support (#51149)

Lyu Han committed 3 months ago

[MISC][Bench] refactor throughput and reuse serve's get samples (#50981)

Jared Wen committed 3 months ago

[Bugfix][LoRA] Guard TrtLlm BF16 MoE LoRA gate on activation type (#51002)

anhtra3889 committed 3 months ago

[Model] Fused mm preprocess normalisation on the Device (#50411)

wang.yuqi committed 3 months ago

[Quantization] Preserve precision in online NVFP4 expert packing (#50029)

Matej Sirovatka committed 3 months ago

[XPU] Register fake meta kernel for fp4_gemm (#50946)

Chaojun Zhang committed 3 months ago

[CPU] Add MLA backend so DeepSeek-V2/V3 can run on CPU (#49453)

maobaolong committed 3 months ago

[GLM Perf] DSv32/glm use skip topk for MTP case, 2.0x kernel performance improvement (#50904)

Wentao Ye committed 3 months ago